test/onboard-selection.test.ts (1)

2075-2094: ⚡ Quick win

Assert probe path behavior explicitly in reasoning mode.

Line [2077] and Line [2150] currently validate end-state only; this test can still pass if /responses (or streaming probe) is called and then fallback succeeds. Please log curl args and assert reasoning mode avoids /responses/-N while still hitting /chat/completions.

Suggested test hardening

@@
     const scriptPath = path.join(tmpDir, "custom-openai-reasoning-check.js");
+    const curlArgsLog = path.join(tmpDir, "custom-openai-reasoning-curl-args.log");
@@
       path.join(fakeBin, "curl"),
       `#!/usr/bin/env bash
+args_log=${JSON.stringify(curlArgsLog)}
+printf '%s\\n' "$*" >> "$args_log"
 body='{"error":{"message":"bad request"}}'
 status="400"
 outfile=""
 url=""
@@
     assert.equal(payload.result.model, "reasoning-model");
     assert.equal(payload.result.preferredInferenceApi, "openai-completions");
     assert.equal(payload.reasoning, "true");
+    const curlInvocations = fs.readFileSync(curlArgsLog, "utf-8");
+    assert.match(curlInvocations, /chat\/completions/);
+    assert.doesNotMatch(curlInvocations, /\/responses/);
+    assert.doesNotMatch(curlInvocations, /(^|\s)-N(\s|$)/);
     assert.ok(
       payload.messages.every(
         (message: string) => !/Enable reasoning mode for this model/.test(message),
       ),
     );

Also applies to: 2150-2160

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@test/onboard-selection.test.ts` around lines 2075 - 2094, The fake curl
script written to path.join(fakeBin, "curl") should record its received
arguments so the test can assert probe behavior: modify the heredoc to append
the full "$@" or each processed arg into a temp log file (e.g.,
"$outfile.args.log" or a known capture file in the test dir) before handling
-o/url logic, and keep returning the same body/status logic; then update the
test assertions to read that captured args log and explicitly assert that in
reasoning mode the captured args do not include '/responses' nor the '-N' flag
and do include a call to '/chat/completions'. Ensure you reference the same
fakeBin/curl generation code and the test's reasoning-mode invocation when
adding the new assertions.

test/onboard.test.ts (1)

391-410: ⚡ Quick win

Add an explicit assertion for the unset/default reasoning path.

This test validates aliases, but not the key default behavior when NEMOCLAW_REASONING is unset. Locking that in will better guard the PR’s intended fallback semantics.

✅ Suggested test addition

   expect(normalizeReasoningFlag("maybe")).toBeNull();

+  delete process.env.NEMOCLAW_REASONING;
+  await expect(configureCompatibleEndpointReasoning()).resolves.toBe("false");
+  expect(process.env.NEMOCLAW_REASONING).toBe("false");
+
   process.env.NEMOCLAW_REASONING = "yes";
   await expect(configureCompatibleEndpointReasoning()).resolves.toBe("true");
   expect(process.env.NEMOCLAW_REASONING).toBe("true");

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@test/onboard.test.ts` around lines 391 - 410, Add an explicit assertion that
the "unset/default" path is covered: before setting NEMOCLAW_REASONING to "yes",
delete process.env.NEMOCLAW_REASONING (or set it to undefined) and call await
expect(configureCompatibleEndpointReasoning()).resolves.toBe(normalizeReasoningFlag(undefined));
this ensures the test asserts the function's fallback behavior when
NEMOCLAW_REASONING is not present and references the existing
normalizeReasoningFlag and configureCompatibleEndpointReasoning symbols.

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

src/lib/state/onboard-session.ts (1)

145-166: ⚡ Quick win

Include compatibleEndpointReasoning in debug summary output.

The new persisted field is not exposed by DebugSessionSummary/summarizeForDebug, which makes reasoning-mode resume issues harder to diagnose from debug dumps.

Proposed patch

 export interface DebugSessionSummary {
   version: number;
   sessionId: string;
   status: string;
   resumable: boolean;
   mode: string;
   startedAt: string;
   updatedAt: string;
   sandboxName: string | null;
   provider: string | null;
   model: string | null;
   endpointUrl: string | null;
   credentialEnv: string | null;
   preferredInferenceApi: string | null;
+  compatibleEndpointReasoning: string | null;
   nimContainer: string | null;
   policyPresets: string[] | null;
   gpuPassthrough: boolean;
   lastStepStarted: string | null;
   lastCompletedStep: string | null;
   failure: SessionFailure | null;
   steps: Record<string, StepState>;
 }
@@
     endpointUrl: redactUrl(session.endpointUrl),
     credentialEnv: session.credentialEnv,
     preferredInferenceApi: session.preferredInferenceApi,
+    compatibleEndpointReasoning: session.compatibleEndpointReasoning,
     nimContainer: session.nimContainer,
     policyPresets: session.policyPresets,

Also applies to: 852-867

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/lib/state/onboard-session.ts` around lines 145 - 166, DebugSessionSummary
and the summarizeForDebug output are missing the new persisted field
compatibleEndpointReasoning, so add compatibleEndpointReasoning: string | null
(matching the persisted type) to the DebugSessionSummary interface and update
the summarizeForDebug function to read the session-compatibleEndpointReasoning
value and include it in the returned summary object (alongside fields like
failure, steps, nimContainer, preferredInferenceApi, etc.) so debug dumps expose
reasoning-mode compatibility; update any related spots mentioned (e.g., the
summarizeForDebug implementation around the lastStep/lastCompletedStep logic) to
populate this field.